Microbiome differences in autism spectrum disorder may reflect dietary preferences that relate to diagnostic features, and we caution against claims that the microbiome has a driving role in ASD

Autism-related dietary preferences mediate autism-gut microbiome associations. Chloe X. Yap et al. Cell, Nov 11 2021. https://doi.org/10.1016/j.cell.2021.10.015

Highlights

• Limited autism-microbiome associations from stool metagenomics of n = 247 children

• Romboutsia timonensis was the only taxa associated with autism diagnosis

• Autistic traits such as restricted interests are associated with less-diverse diet

• Less-diverse diet, in turn, is associated with lower microbiome alpha-diversity

Summary: There is increasing interest in the potential contribution of the gut microbiome to autism spectrum disorder (ASD). However, previous studies have been underpowered and have not been designed to address potential confounding factors in a comprehensive way. We performed a large autism stool metagenomics study (n = 247) based on participants from the Australian Autism Biobank and the Queensland Twin Adolescent Brain project. We found negligible direct associations between ASD diagnosis and the gut microbiome. Instead, our data support a model whereby ASD-related restricted interests are associated with less-diverse diet, and in turn reduced microbial taxonomic diversity and looser stool consistency. In contrast to ASD diagnosis, our dataset was well powered to detect microbiome associations with traits such as age, dietary intake, and stool consistency. Overall, microbiome differences in ASD may reflect dietary preferences that relate to diagnostic features, and we caution against claims that the microbiome has a driving role in ASD.

Discussion

In this large ASD stool metagenomics study in which confounders were carefully considered, we found negligible evidence for direct associations between the stool microbiome and ASD diagnostic status, which was also the case for other neurodevelopmental traits (e.g., IQ-DQ, sleep problems). For ASD, there was limited evidence for associations with taxonomic diversity or microbiome-association index (b2; Figure 2), and only one differentially abundant species was robustly identified (Figure 3). These results were striking when compared to strong associations of microbiome composition with age, diet, and stool consistency (Figure 2). Importantly, we failed to replicate previously reported ASD-microbiome associations from human studies. Instead, we found evidence linking behaviors associated with the autism spectrum (e.g., repetitive-restricted behaviors or interests, sensory preferences, and social affect) to reduced dietary diversity, which, in turn, was associated with reduced microbiome diversity and looser stool consistency (Figure 4J). This putative model challenges suggestions from animal studies that the microbiome may be causally related to ASD-related traits (

Buffington et al., 2016

; 

Hsiao et al., 2013

; 

Sharon et al., 2019

). Our findings also stand at odds to the proliferation of experimental interventions and early clinical trials that propose to “treat” ASD by targeting the microbiome (

Kang et al., 2019

; 

Kang et al., 2017

).

In contrast to measures of microbiome composition, ASD was robustly and significantly linked to dietary variables, irrespective of covariates (Table S3). We found (1) that significant variance in ASD diagnosis was associated with diet but not the microbiome in the b2 analysis (Figure 2), (2) reduced meat intake in the ASD group (Figure S5), and (3) reduced dietary diversity in the ASD group despite significantly higher variance in dietary diversity (Figure 4A), which is consistent with the dietetics literature (

Panossian et al., 2020

; 

Schreck and Williams, 2006

) and some smaller ASD microbiome studies with dietary data (

Berding and Donovan, 2018

).

One rationale for the interest in the ASD microbiome is the frequent co-occurrence of gastrointestinal complaints (

Chaidez et al., 2014

; 

Kohane et al., 2012

; 

McElhanon et al., 2014

). In the absence of complete gastrointestinal complaint reporting, we analyzed stool consistency scores, with the caveat that it is unclear how this single-time point data reflects chronic conditions. Stool consistency appeared to be more proximal to taxonomic than dietary diversity, although we acknowledge that it is difficult to distinguish between a top-down (i.e., dietary and taxonomic diversity influencing downstream stool consistency) versus bottom-up (i.e., stool consistency being an upstream proxy) relationship. For the former, dietary restrictedness could plausibly affect gut ecology and taxonomic diversity, which in turn affects stool consistency. In relation to a bottom-up model, looser stool may indicate underlying food allergies or intolerances, which may be associated with deliberate (parental) dietary restriction to identify causative agents. Additionally, looser stool consistency reflects reduced gastrointestinal transit time and reduced colonic water reuptake (

Vandeputte et al., 2016

), which affects taxonomic diversity. As the narrow-sense heritability of gastrointestinal conditions that affect stool consistency (e.g., irritable bowel syndrome) are small (

Wu et al., 2021

), environmental contributions likely predominate over genetics (

Rothschild et al., 2018

).

Our results have important implications for understanding the role of the gut microbiome in ASD and other psychiatric traits. First, in relation to medical care, food selectivity among children on the autism spectrum is an important clinical concern. Food selectivity is related to avoidant/restrictive food intake disorder (ARFID; which is likely underdiagnosed despite affecting over 20% of autistic children [

Koomar et al., 2021

]) and can cause nutritional deficiencies among autistic children (

Zimmer et al., 2012

) to the extent that hospitalization and invasive measures such as enteral feeding are required (

Tang et al., 2011

). Our results also suggest that dietary quality is poorer in children on the spectrum (Methods S1). Given that elevated microbial diversity is robustly associated with improved health outcomes (

Valdes et al., 2018

), the association of ASD with poorer dietary quality and reduced dietary and taxonomic diversity underlines the importance of dietary and nutritional interventions in this population. Second, our results have implications for the interpretation of cause and effect in relation to diet in microbiome analyses in psychiatric conditions. There is growing interest in the contribution of diet and the microbiome to psychiatric traits (e.g., depression [

Dash et al., 2015

; 

Molendijk et al., 2018

]), but our results emphasize the need to consider the (arguably more intuitive) impact of behavior on the microbiome (

Jacka et al., 2015

). These results add to other reports of diet driving microbiome associations with health (

Claesson et al., 2012

).

For future microbiome studies, we emphasize the importance of collecting detailed dietary data (recent examples [

Asnicar et al., 2021

; 

Wang et al., 2021

]), particularly for ASD and other neuropsychiatric traits with plausible co-variation of diet with diagnosis or treatment. We advocate for larger sample sizes to ensure that results are robust to sampling effects and to identify subtler microbiome associations. We also recommend higher-resolution metagenomics technology and expanded databases since more granular taxonomic measures of microbiome composition were more sensitive (Table S1), gene-level ORMs explained more variance for some traits (Table S1), power to detect associations was weaker with the MetaPhlAn2/NCBI pipeline (Methods S1), and because taxonomic and functional datasets may capture complementary aspects of the microbiome (Figures S1 and S3).

In conclusion, we found negligible direct associations between ASD and the gut microbiome in contrast to strong associations with other phenotypes such as age, dietary variables, and stool consistency. Instead, we find evidence that restricted dietary diversity and poorer quality—which is associated with specific ASD features such as restrictive-repetitive behaviors—is a significant mediator of taxonomic diversity, and in turn, stool consistency. Our results are consistent with an upstream role for ASD-related behaviors and dietary preferences on the gut microbiome and are contrary to claims of the microbiome having a major (or causal) role in ASD.

 Limitations of the study

First, this study did not have a longitudinal design, so we cannot rule out microbiome contributions prior to ASD diagnosis. Second, although this is to our knowledge the largest metagenomics study of the ASD stool microbiome to date, there may nonetheless be sampling biases that require larger studies to overcome (

Rothschild et al., 2020

). Third, this study used stool samples as a gut microbiome proxy, which may not accurately represent the more difficult-to-access mucosal microbiome (

Shanahan et al., 2016

). Fourth, data on antibiotic intake in this cohort were not systematically collected and so could not be rigorously accounted for other than through exclusion in sensitivity analyses. Fifth, the gold-standard differential abundance analysis relied on per-feature tests that do not reflect the interactions and non-independence that occurs within an ecological or metabolic context. Finally, we await the emergence of datasets with comparable study design, consideration of confounders, and depth of phenotypic and metagenomics data for replication of these results.